Security is a Holistic Proposition

Gorka Sadowski

Subscribe to Gorka Sadowski: eMailAlertsEmail Alerts
Get Gorka Sadowski: homepageHomepage mobileMobile rssRSS facebookFacebook twitterTwitter linkedinLinkedIn

Related Topics: SOA & WOA Magazine, Java Developer Magazine


Why Rule-Based Log Correlation Is Almost a Good Idea - Part 3

Some examples of scenario that many people like to brag about

We'll see below some examples of security attack scenario that many people will put forth as a perfect example of how powerful, valuable and simple correlation is.

As you can see, the overall approach of using static rule-based correlation on these is simply flawed.

Attack Scenario Example 1: Identity Theft
There are numerous ways to perform an Identity Theft attack, but let's focus on just one of them, recognizing that somebody cannot be in two places at the same time and hence that a user cannot log in your infrastructure from VPN and locally from the office "at the same time." Furthermore, if he connects through VPN, then disconnects and then "shortly thereafter" he reconnects locally, then it is probably Identity Theft.

A scenario might be:

  • If one of my users is logging in my infrastructure from the Internet through my VPN, then logs out.
  • And then some time later that same user logs in locally inside the office.
  • Then there is "probably" an identity theft happening and I need to ring an alert.

Proper Risk Management methodology implies that this "probable identity theft" needs to be quantified so as to prioritize risks and understand the best way to deal with it. In order to do this quantification, several factors need to be taken into account:

- How long is "some time later"?

o Is five minutes good? In other words if he logs in/out the VPN and then less than five minutes later he logs in locally, then is there identity theft?

§ But then what happens if the bad guy logged in/out the VPN one hour before the user logs in locally?

The attack will not be detected.

o Is one hour good?

§ What happens if the bad guy logged in/out the VPN two hours before the user logs in locally?

The attack will not be detected.

o Is two hours good?

§ Meaning that if he logs in via VPN and then two hours later he logs in/out locally then is there identity theft?

What happens if he lives next door to his work and decides to check his email while having breakfast and then an hour later he's at work?

This will generate a false positive.

So maybe we need to tie where he lives with where his office is?

§ In this case, we´ll give him "reasonable time" to get to work, and ring an alert if timing is off-base compared to his commute time.

§ What about if he checks his email from his favorite café on his way to work? Then the time difference between connection through VPN or local login will be shorter than expected.

This will generate a false positive.

§ What about if it's Friday, the day when there is that bad traffic jam on the highway?

A potential attack will not be detected.

§ And how about "there was an accident on the way here, traffic was terrible"

A potential attack will not be detected.

So then maybe we need to tie where he's logging from when using VPN with where the office is located

§ That way if he logs in from VPN 1000 miles from the office and then logs in the office, say five hours later then it's flagged as identity theft

§ But if what if he checks his email in the airport terminal before embarking on a flight and goes to work straight after arriving?

This will generate a false positive

So, what if...?

And what happens when...?

It's Friday but the first Friday of January, it's still vacation, it's a bissextile year, but... so the rule should be...?

And this user lives this far from the office... but he often rides his bicycle to work... except when it rains... so the rule should be...?

For each user!!!

Have we progressed in the quantification of "probably an identity theft"? Not much, but now we are left managing dozens of variations of correlation rules and having to manage dozens of exceptions and complex geo-localization information rules and weather reports, and time of the day information for just one simple attack scenario.

And still many (most?) attacks are not detected... But you are nonetheless left with plenty false positives.

Keeping state information on potentially thousands of connections, through potentially several hours will quickly exhaust your correlation engine resources, requiring you to throw lots of hardware to your correlation engine so that it doesn't blow up after a few minutes.

Keep in mind that if your user is first logging in/out locally and then through the VPN, then all of this needs to be reprogrammed. Likewise if your user is coming through SSL VPN instead of IPSec VPN. How about if he logs into the Intranet from outside while he was just here locally. Or vice versa. And...

Imagine complex attack scenarios...

Doesn't work, doesn't scale and carries a very high operational Total Cost of Ownership.

More Stories By Gorka Sadowski

Gorka is a natural born entrepreneur with a deep understanding of Technology, IT Security and how these create value in the Marketplace. He is today offering innovative European startups the opportunity to benefit from the Silicon Valley ecosystem accelerators. Gorka spent the last 20 years initiating, building and growing businesses that provide technology solutions to the Industry. From General Manager Spain, Italy and Portugal for LogLogic, defining Next Generation Log Management and Security Forensics, to Director Unisys France, bringing Cloud Security service offerings to the market, from Director of Emerging Technologies at NetScreen, defining Next Generation Firewall, to Director of Performance Engineering at INS, removing WAN and Internet bottlenecks, Gorka has always been involved in innovative Technology and IT Security solutions, creating successful Business Units within established Groups and helping launch breakthrough startups such as KOLA Kids OnLine America, a social network for safe computing for children, SourceFire, a leading network security solution provider, or Ibixis, a boutique European business accelerator.